MR has undeniably emerged as a powerful tool for addressing causal questions in observational research (Fig. 1c); however, the abundance of MR studies over the past three years varies in quality and relevance, hindering clear interpretation and limiting the applicability of current MR findings. This is a common concern across various fields, and the field of osteoporosis research is no exception. Recently, Burgess et al. [36] provided a comprehensive review of the common pitfalls in conducting a reliable MR analysis. The first highlighted key issue is evaluating inappropriate research questions (“Is the research question addressable using MR?”). It’s important to have a strong logical rationale for why a research question warrants further exploration. The initial MR studies have made significant strides toward clarifying whether the associations observed in cross-sectional studies are truly causal or confounded by other variables [37]. One significant advancement in the field was the development of two-sample MR, which proved highly beneficial for studying associations between risk factors and outcomes in settings where both the exposure and outcome were not measured in the same study. Additionally, this approach helped improve statistical power. However, it also led to an increase in the number of studies, some of which are valuable but introduced certain challenges.

Over the past three years, up to 265 MR studies have been conducted in the bone field, out of which 229 studies focused on osteoporosis (n = 117), fracture risk (n = 39), and BMD (n = 164) as main outcomes, covering risk factors across 32 domains (Fig. 2, Supplementary Table 1). 32 of these studies (n = 229) perform bidirectional MR, in which bone phenotypes serve as both exposures and outcomes, typically used when the direction of effect is uncertain (Supplementary Table 2). The remaining 29 studies studied BMD as the main exposure for various outcomes (Supplementary Table 3).

Summary of 32 risk factor domains tested for associations with bone outcomes. n: number of studies that have evaluated the particular trait as exposure for MR analysis using bone outcomes; COPD: chronic obstructive pulmonary disease; NAFLD: non-alcoholic fatty liver disease; GERD: gastroesophageal reflux disease; FGF: fibroblast growth factor; sRANKL: soluble receptor activator of NF-κB ligand; OPG: osteoprotegerin; BMI: body mass index; WHR: waist-to-hip ratio; WC: waist circumference; HC: hip circumference

A chronic condition that has been extensively evaluated in the last few years is non-alcoholic fatty liver disease (NAFLD). Zhou et al. reported a causal association between genetically predicted NAFLD and osteoporosis [38]. Similarly, Pei et al. [39] identified a causal relationship between NAFLD and femoral neck BMD (FN-BMD), along with a suggestive association between NAFLD and osteoporosis, which was also supported by findings from Cui et al. [40]. In addition, a causal link between NAFLD and forearm BMD and total body BMD has been reported [41, 42]. In contrast, Huang et al. found no significant effects of NAFLD on BMD in four skeletal sites (total body, femoral neck, lumbar spine, forearm) [43]. When Pei et al. [39] used the same data source (FinnGen) to derive SNPs for the NAFLD instrumental variable as Huang et al. [43], they similarly reported a null effect on BMD. This highlights the importance of carefully considering differences in instrumental variable construction when comparing findings across studies. A causal relationship between Crohn’s disease [44], inflammatory bowel disease [44,45,46,47], ulcerative colitis [44], and osteoporosis has also been reported by several MR studies.

Next, MR studies have also evaluated the role of COPD and Asthma in the development of osteoporosis [48, 49]. Osteoporosis and vertebral fractures are quite common in patients with advanced COPD and show a significant relationship to the mortality of these patients [50, 51]. A causal relationship between COPD and the risk of osteoporosis has been reported by Dou et al. [52], along with evidence that COPD is negatively associated with heel BMD [48] but no other site-specific BMDs. However, findings from Yang et al. did not support a causal relationship between COPD and heel BMD [49]. The discrepancy between the two studies likely arises from the differences in the construction of the instrumental variable.

Furthermore, MR has reported that both childhood-onset asthma and adult-onset asthma [53] may have a causal effect on osteoporosis, with a higher risk of osteoporosis observed in adult-onset asthma. Nevertheless, the results need to be interpreted with caution, as the effect estimate was not robust across different MR estimators. MR, like any other statistical approach, is subject to limitations. To address potential biases and strengthen causal inference, it is common practice to test a range of MR estimators, including inverse variance weighting (IVW), the weighted median, and mode-based estimators, to evaluate whether the causal effect estimates are consistent across methods that rely on different assumptions about the validity of the instrumental variable.

An MR study has suggested that genetically predicted calcium channel blockers (CCBs) may increase fracture risk, whereas angiotensin receptor blockers (ARBs) may decrease fracture risk [54]. No effect was reported for alpha-blockers, angiotensin-converting enzyme inhibitors (ACEIs), beta-blockers (BBs), loop diuretics, potassium-sparing diuretics (PSDs), and thiazide diuretics. The findings on ARBs, broadly classified as Renin-angiotensin-aldosterone system (RAAS) inhibitors, are particularly interesting, as animal studies have demonstrated increased bone mass and strength through the blockade of angiotensin II pathways [55]. Nevertheless, similar effects have not been observed with other classes of RAAS inhibitors, specifically the angiotensin-converting enzyme (ACE) inhibitors.

In similar study settings, two MR studies have reported conflicting effects of PCSK9 inhibitors on osteoporosis risk. These inhibitors are monoclonal antibodies that decrease LDL cholesterol by blocking PCSK9, a protein that typically promotes the breakdown of LDL receptors. PCSK9 inhibitors play a key role in managing cardiovascular disease, and it is hypothesized that they could affect bone metabolism by lowering LDL levels [56]. However, evidence from these two MR studies remains inconsistent: one two-sample MR study reported a lower osteoporosis risk [57], whereas another reported an increased osteoporosis risk [58]. One difference between the studies lies in the SNPs used to construct the instrumental variable (assuming there are no other technical biases), which brings us to the second key consideration in evaluating the credibility of an MR study: “Are the chosen genetic variants appropriate?” [36]. MR analysis has also suggested a potentially beneficial effect of another class of lipid-lowering medication, namely the peroxisome proliferator-activated receptor (PPARs) agonists (bezafibrate and fenofibric acid) [59]. The drug-target MR approach analysis has also identified potential novel drug targets, suggesting that ANGPTL3 and APOC3 may serve as new non-statin lipid-lowering drugs for treating or preventing osteoporosis.

Advancements in proteomics have significantly transformed the MR field, enabling the identification of potential causal proteins and druggable targets for diseases, including osteoporosis. Up to 13 studies have used proteomics as an exposure, with 11 performing protein-wide association analyses [60,61,62,63,64,65,66,67,68,69,70,71,72], and one has focused on serum ADAM/ADAMTS levels [60], and another on mTOR-dependent EIF-4E circulating protein levels [70]. The latter studies have shown ADAM/ADAMTS levels to have a suggestive impact on estimated heel BMD and no effect on femoral neck, lumbar spine, and forearm BMD, whereas one mTOR-dependent circulating protein influenced forearm BMD but not the other skeletal sites. The relevance of these proteins may be limited as the site-specific effects are unclear. One interesting study is that by Zhou et al. [71], where integrating exome sequencing findings with proteomics MR evidence has prioritized CD109 (cluster of differentiation 109) as a potential novel drug target for osteoporosis. The exome analysis revealed that heterozygous loss-of-function variants in CD109 are associated with increased BMD, and MR analysis further supported this, showing that lower circulating CD109 levels correlate with higher estimated heel BMD. Subsequent functional experiments confirmed that partial CD109 knockdown was associated with increased bone mineralization, reinforcing its therapeutic potential. Next, Michaelsson et al. [72] have identified 24 cardiometabolic proteins associated with fracture risk, of which SOST, NTproBNP, BNP, and CCDC80 have been demonstrated to have potential causal effects. Regarding CCDC80, the observational study and the MR analysis showed opposite effects, which the authors hypothesize could be due to horizontal pleiotropy. This study is a good example of how bringing together evidence from multiple, complementary study settings, so-called triangulation, can help us better understand the findings and strengthen the validity of MR findings.

Up to six studies [73,74,75,76,77,78] have been performed in the metabolomics domain in the last three years. A large-scale plasma metabolite study by Chen et al. [73] showed that orotate has an adverse effect on bone, as it is causally associated with lower estimated heel BMD and, in a separate independent cohort, with increased fracture risk. Next, Liu et al. [74] identified five metabolites that may contribute to osteopenia by combining SMR and colocalization analyses. Among these, biliverdin appeared to have a protective effect; however, the implications are unclear. Chen et al. [77] found 22, 10, 3, 7, and 2 metabolite links to estimated heel BMD, total body BMD, lumbar spine BMD, femoral neck BMD, and forearm BMD, respectively, with androsterone sulfate showing a strong effect across all site-specific BMDs. However, in this study, the authors used odds ratios to report continuous outcomes, which complicates the interpretation of the effect’s direction. Thus, authors need to report the risk estimates on the appropriate risk scale. Next, Gong et al. [76] report a causal association between the Adenosine 5’-monophosphate (AMP) to alanine ratio and BMD. However, the metabolite GWAS source cited for the metabolomics MR analysis does not include a measurement of the adenosine 5’-monophosphate (AMP) to alanine ratio, making it unclear or implausible to construct the instrumental variable. Moreover, this specific ratio is not commonly used in clinical or research practice, which further limits the interpretability of the findings. Therefore, caution is warranted when deriving conclusions from MR analyses if the construction of the instrumental variable is not clearly reported.

Finally, although it is less likely that SNPs associated with microbiome diversity could serve as a plausible exposure for MR [38], up to 15 MR studies have been conducted on microbiota [69, 78,79,80,81,82,83,84,85,86,87,88,89,90,91] in the past three years. Among protective taxa, Coprococcus (especially Coprococcus2/3) has consistently emerged, showing inverse associations with OP and positive associations with BMD across multiple independent cohorts (Chen et al. [88], Qiao et al. [80], Zhou et al. [81], Ma et al. [83], Wang et al. [89]). Similarly, Prevotellacea frequently demonstrated beneficial effects on BMD [86, 88], whereas Burkholderiales showed a negative association with OP and decreased osteoclast activation [83, 85, 87]. On the other hand, multiple Ruminococcaceae sub-groups have been implicated in decreased BMD and higher OP risk across studies (Chen et al. [88], Qiao et al. [80], Xue et al. [86], Wang et al. [89]). Peptococcaceae and Desulfobacterota/Desulfovibrionaceae were also repeatedly associated with BMD and osteoporosis [79, 88].

Romosozumab, a monoclonal antibody that targets sclerostin, is an effective treatment for osteoporosis in people at high fracture risk. However, two phase III randomized controlled trials [92, 93] have reported an increased incidence of adverse cardiovascular events in patients receiving romosozumab. In contrast, another randomized controlled trial did not observe this adverse effect [94]. An MR study [95] has shown that genetically lower sclerostin levels are associated with increased risks of hypertension, myocardial infarction, type 2 diabetes, and greater coronary artery calcification. However, a novel flexible machine learning MR approach found no effect of sclerostin on ischemic cardiovascular diseases [96]. Although the impact remains uncertain, drug agencies have been cautious and have issued contraindications for patients with a history of myocardial infarction or stroke.